George Miller's 1956 paper "The Magical Number Seven, Plus or Minus Two" is one of the most cited papers in the history of psychology. Miller reviewed experiments in which observers made absolute judgments about unidimensional stimuli — tones varying in pitch, visual stimuli varying in brightness, salt solutions varying in concentration — and found a remarkably consistent result: information transmission plateaued at about 2.5 bits, corresponding to roughly 5 to 9 discriminable categories.
Channel Capacity for Absolute Judgment
T_max ≈ 2.5 bits for unidimensional stimuli
≈ log₂(6) ≈ 5–9 discriminable categories
Multidimensional stimuli: T increases but < sum of individual T values
The information-theoretic analysis treats the observer as a communication channel. The input is the stimulus set with entropy H(S), the output is the response set, and the channel capacity is the maximum mutual information between stimulus and response. Garner and Hake (1951) pioneered this analysis, showing that for absolute judgments of loudness, information transmission saturates even as the number of stimulus categories increases far beyond 7. The bottleneck is not in perception — observers can detect very small differences — but in the labeling process that maps continuous sensory experience onto discrete categories.
Recoding and Chunking
Miller's most influential contribution was the concept of chunking: the process of recoding information into higher-order units. Although the channel capacity for individual items is limited, the capacity for chunks is approximately constant. A chess master, for example, does not remember individual piece positions but rather familiar configurations — each chunk carries far more information than a single piece position. Miller argued that recoding is the fundamental cognitive strategy for overcoming capacity limitations.
Subsequent research by Cowan (2001) argued that the true capacity of working memory is closer to 4 chunks, not 7, once rehearsal and grouping strategies are controlled. The discrepancy arises because Miller's original estimate conflated storage capacity (the number of chunks that can be maintained) with the information content per chunk (which varies with expertise and recoding). The limit of 2.5 bits for absolute judgment remains well established, but the "magical number" for working memory continues to be debated.
Multidimensional Extension
When stimuli vary on multiple dimensions simultaneously (e.g., pitch and loudness), information transmission increases substantially — observers can identify more stimulus categories. However, the increase is less than the sum of the individual channel capacities, indicating that the dimensions are not processed with full independence. Garner (1962) systematically investigated these multidimensional effects, distinguishing between integral dimensions (which interact) and separable dimensions (which are processed more independently).
Miller's information-theoretic analysis of human performance established a productive bridge between Shannon's engineering framework and cognitive psychology. The concept of a processing bottleneck, characterized in bits, continues to influence theories of attention, working memory, and perceptual categorization.